Soft phenotyping for sepsis via EHR time-aware soft clustering

Area of Science:

• Critical care medicine
• Computational biology
• Health informatics

Background:

• Sepsis is a life-threatening condition with high mortality, stemming from a dysregulated immune response to infection.
• Identifying sepsis sub-phenotypes is crucial for understanding disease variability, optimizing treatments, and improving patient outcomes.
• Previous sub-phenotyping methods using electronic health records (EHRs) lacked temporal information and made uncertain assumptions.

Purpose of the Study:

• To develop a novel, time-aware clustering algorithm for identifying sepsis sub-phenotypes.
• To utilize clinical variables from EHR data for more accurate sepsis characterization.
• To advance the understanding of sepsis heterogeneity and improve clinical decision-making.

Main Methods:

• Developed a time-aware soft clustering algorithm incorporating clinical variables.
• Applied the algorithm to electronic health record (EHR) data to identify sepsis sub-phenotypes.
• Evaluated the medical plausibility of the identified sub-phenotypes and developed an early-warning prediction model.

Main Results:

• Identified six novel hybrid sepsis sub-phenotypes with clinical plausibility.
• Demonstrated the algorithm's ability to capture temporal dynamics in sepsis.
• Developed a logistic regression model for early sepsis prediction.

Conclusions:

• The novel sepsis hybrid sub-phenotypes offer more accurate insights into organ dysfunction and recovery.
• These findings can inform sepsis management decisions and improve patient prognosis.
• The time-aware approach enhances the characterization of sepsis heterogeneity.